Tapping dust removal control method and device based on multi-mode fusion sensing intelligent agent

Abstract

The embodiment of the invention provides a tapping dust removal control method and device based on a multi-mode fusion sensing intelligent agent, and the method comprises the steps: collecting the environment state parameters of a tapping site and the motion parameters of dust removal equipment according to a fixed time interval; performing feature extraction on the environment state parameters and the dust removal equipment action parameters, and constructing an evaluation network and a target network based on feature values; based on a deep reinforcement learning algorithm, training a reinforcement learning process of interaction between the intelligent agent and the environment through the evaluation network and the target network, and obtaining an optimization model of a tapping dust removal control strategy; the optimization model is applied to the dust removal equipment so as to carry out dust removal control on the tapping site; real-time monitoring of working condition changes and smoke dust concentration conditions in the blast furnace tapping process and intelligent control of the dust removal equipment can be achieved, the control strategy of the dust removal equipment is automatically adjusted, the dust removal effect is improved, and the energy consumption of the equipment is reduced.

Description

Technical Field

[0001] The present application relates to the field of dust removal control, and specifically to a method and device for iron tapping dust removal control based on a multimodal fusion perception intelligent agent. Background Art

[0002] With the rapid development of industrial production, the problem of smoke pollution generated by the blast furnace tapping process in the steel industry has become increasingly serious. In order to reduce smoke emissions and improve environmental performance, dust removal equipment is widely used in the blast furnace tapping process.

[0003] However, traditional dust removal control strategies have many problems.

[0004] First, because the dust collector must be configured and operated by a human operator, a large number of workers are required to monitor the dust removal process, which undoubtedly results in high additional labor costs. However, the current development trend of metallurgical plants is shifting towards a management model with no or minimal human operators.

[0005] Furthermore, because the control of the iron-making equipment often relies on the immediate reactions and judgment of workers, implementing appropriate dust removal measures also requires a certain amount of time. This reliance on manual labor can easily lead to underestimation of the situation, which can lead to untimely or excessive dust removal, resulting in unnecessary energy waste.

[0006] In addition, the power and speed regulation of dust removal equipment often requires empirical determination and uses fixed discrete parameter settings, which cannot be dynamically and linearly adjusted according to actual working conditions.

[0007] Finally, most factories tend to have relatively simple methods of measuring smoke and dust; some companies have multiple detection methods, but are unable to effectively integrate the data from these multiple sources to make more scientific judgments, and still only use single-source data for independent analysis. The value of the data urgently needs to be more fully explored. Summary of the Invention

[0008] In response to the problems in the existing technology, the present application provides a method and device for controlling iron-making and dust removal based on a multimodal fusion perception intelligent agent. It can use multimodal fusion perception intelligent agent technology to realize real-time monitoring of operating condition changes and smoke dust concentration during blast furnace iron-making and intelligent control of dust removal equipment. The intelligent agent constructed through deep reinforcement learning algorithm can automatically adjust the control strategy of dust removal equipment, improve dust removal effect and reduce equipment energy consumption.

[0009] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0010] According to a first aspect of an embodiment of the present application, the present application provides a method for controlling dust removal during iron tapping based on a multimodal fusion perception agent, comprising:

[0011] Collect environmental parameters of the iron-making site and dust removal equipment operating parameters at fixed time intervals;

[0012] Extracting features of the environmental state parameters and the dust removal equipment action parameters, and constructing an evaluation network and a target network based on the feature values;

[0013] Based on the deep reinforcement learning algorithm, an optimization model of the iron tapping dust removal control strategy is obtained through the reinforcement learning process of the evaluation network and the target network training agent interacting with the environment;

[0014] The optimization model is applied to the dust removal equipment to perform dust removal control on the iron-making site.

[0015] According to any embodiment of the present application, the environmental state parameters collected at the iron-making site include:

[0016] Use the camera to collect smoke and dust images of the current iron-making environment;

[0017] The smoke data of the current iron-making environment is collected through the total suspended particulate matter sensor and laser sensor.

[0018] According to any embodiment of the present application, the operating parameters of the dust removal equipment at the iron-making site include the dust removal fan speed and the valve opening.

[0019] According to any embodiment of the present application, the feature extraction of the environmental state parameters and the dust removal equipment action parameters includes:

[0020] performing feature extraction on the smoke image based on an image encoder;

[0021] Constructing a complete text description based on the smoke data, extracting features from the text description using a text encoder, and dimensionally fusing features of the smoke image with features of the smoke data;

[0022] Feature extraction is performed on the action parameters of the dust removal equipment based on the text encoder.

[0023] According to any embodiment of the present application, the reinforcement learning process of training the agent to interact with the environment through the evaluation network and the target network based on the deep reinforcement learning algorithm includes:

[0024] Initializing the evaluation network and the target network so that the evaluation network and the target network have consistent structures and independent parameters;

[0025] In each training cycle, an experience replay mechanism is used to randomly sample from historical interactions to reduce correlation and variance in training;

[0026] Based on the temporal difference update rule, the weights of the evaluation network are adjusted after each training cycle according to the reward signal and the maximum expected benefit of the next state to select the execution action that maximizes the expected reward.

[0027] According to any embodiment of the present application, it also includes:

[0028] A reward function is constructed with the maximum dust removal effect and the minimum energy consumption of the dust removal equipment as reward indicators, where:

[0029] In response to a decrease in smoke concentration detected by the sensor, the reward increases;

[0030] In response to energy consumption resulting from operation of the device exceeding a predetermined threshold, the reward is reduced.

[0031] According to any embodiment of the present application, it also includes:

[0032] In response to the agent decision process starting, setting an exploration rate value, and selecting a random action from the entire set of actions at the exploration rate;

[0033] The exploration rate is lowered during training to reduce random exploration behavior, and the exploration rate reduction strategy is determined based on the learning progress of the agent and environmental feedback.

[0034] According to any embodiment of the present application, applying the optimization model to the dust removal equipment to perform dust removal control on the iron tapping site includes:

[0035] Dynamically adjusting the action parameters of the dust removal equipment according to the control strategy output by the optimization model;

[0036] The overall performance of the dust removal system is evaluated regularly. If it is detected that the dust removal effect and equipment energy consumption do not meet the expected indicators, feedback will be provided to the next reinforcement learning process.

[0037] According to a second aspect of the embodiments of the present application, the present application provides a dust removal control device for iron tapping based on a multimodal fusion perception agent, comprising:

[0038] The multimodal environmental perception module is used to collect environmental parameters of the ironworks site and the operating parameters of the dust removal equipment at fixed time intervals.

[0039] A feature extraction module is used to extract features of the environmental state parameters and the dust removal equipment action parameters, and to construct an evaluation network and a target network based on the feature values;

[0040] An agent learning module is used to: based on a deep reinforcement learning algorithm, train the reinforcement learning process of the agent's interaction with the environment through the evaluation network and the target network to obtain an optimization model of the iron tapping dust removal control strategy;

[0041] The intelligent agent decision module is used to apply the optimization model to the dust removal equipment to perform dust removal control on the iron-making site.

[0042] According to any embodiment of the present application, the agent learning module includes:

[0043] The image acquisition unit is used to: acquire smoke and dust images of the current iron-making environment through a camera;

[0044] The data acquisition unit is used to collect smoke data of the current iron-making environment through a total suspended particulate matter sensor and a laser sensor.

[0045] According to any embodiment of the present application, the operating parameters of the dust removal equipment at the iron-making site include the dust removal fan speed and the valve opening.

[0046] According to any embodiment of the present application, the feature extraction module includes:

[0047] An image feature unit, configured to extract features from the smoke image based on an image encoder;

[0048] a data feature unit, configured to construct a complete text description based on the smoke data, extract features of the text description using a text encoder, and perform dimension fusion on features of the smoke image and features of the smoke data;

[0049] An action feature unit is used to extract features of the action parameters of the dust removal equipment based on the text encoder.

[0050] According to any embodiment of the present application, the agent learning module includes:

[0051] An initialization unit, configured to initialize the evaluation network and the target network so that the evaluation network and the target network have consistent structures and independent parameters;

[0052] The training unit is used to: randomly sample from historical interactions using the experience replay mechanism in each training cycle to reduce the correlation and variance in training;

[0053] The optimization unit is used to adjust the weights of the evaluation network after each training cycle based on the reward signal and the maximum expected benefit of the next state based on the temporal difference update rule to select the execution action that can maximize the expected reward.

[0054] According to any embodiment of the present application, it further includes a reward building module for:

[0055] A reward function is constructed with the maximum dust removal effect and the minimum energy consumption of the dust removal equipment as reward indicators, where:

[0056] In response to a decrease in smoke concentration detected by the sensor, the reward increases;

[0057] In response to energy consumption resulting from operation of the device exceeding a predetermined threshold, the reward is reduced.

[0058] According to any embodiment of the present application, a greedy optimization module is further included, including:

[0059] an exploration selection unit, configured to: in response to the agent decision process starting, set an exploration rate value, and select a random action from the entire action set at the exploration rate;

[0060] The exploration optimization unit is used to reduce the exploration rate during training to reduce random exploration behavior, and determine the exploration rate reduction strategy based on the learning progress of the agent and environmental feedback.

[0061] According to any embodiment of the present application, the agent decision module includes:

[0062] A control output unit, configured to dynamically adjust the action parameters of the dust removal equipment according to the control strategy output by the optimization model;

[0063] The feedback improvement unit is used to regularly evaluate the overall performance of the dust removal system. If it is detected that the dust removal effect and equipment energy consumption do not meet the expected indicators, feedback will be provided to the next reinforcement learning process.

[0064] According to the third aspect of the embodiment of the present application, the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the steps of the iron-out dust removal control method based on a multimodal fusion perception intelligent agent are implemented.

[0065] According to the fourth aspect of the embodiments of the present application, the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the iron-out dust removal control method based on a multimodal fusion perception intelligent agent are implemented.

[0066] According to the fifth aspect of the embodiment of the present application, the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the iron-out dust removal control method based on a multimodal fusion perception intelligent agent.

[0067] It can be seen from the above technical solution that the present application provides a method and device for controlling dust removal during iron tapping based on a multimodal fusion perception intelligent agent, which collects environmental state parameters and dust removal equipment action parameters of the iron tapping site according to fixed time intervals; extracts features of the environmental state parameters and the dust removal equipment action parameters, and constructs an evaluation network and a target network based on the feature values; based on a deep reinforcement learning algorithm, the evaluation network and the target network are used to train the reinforcement learning process of the interaction between the intelligent agent and the environment to obtain an optimization model of the iron tapping dust removal control strategy; the optimization model is applied to the dust removal equipment to perform dust removal control on the iron tapping site; the multimodal fusion perception intelligent agent technology can be used to realize real-time monitoring of operating condition changes and smoke concentration conditions during blast furnace iron tapping and intelligent control of dust removal equipment. The intelligent agent constructed by the deep reinforcement learning algorithm can automatically adjust the control strategy of the dust removal equipment, improve the dust removal effect and reduce equipment energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0069] Figure 1 This is one of the flow charts of the iron tapping dust removal control method based on a multimodal fusion perception agent in an embodiment of the present application;

[0070] Figure 2 This is a second flow chart of the iron tapping dust removal control method based on a multimodal fusion perception agent in an embodiment of the present application;

[0071] Figure 3 This is a third flow chart of the iron tapping dust removal control method based on a multimodal fusion perception agent in an embodiment of the present application;

[0072] Figure 4 Schematic diagram of the Q-function network structure based on a multimodal fusion perception agent in an embodiment of the present application;

[0073] Figure 5 This is a fourth flow chart of the iron tapping dust removal control method based on a multimodal fusion perception agent in an embodiment of the present application;

[0074] Figure 6 Schematic diagram of the training process of the iron tapping dust removal control method based on the multimodal fusion perception agent in the embodiment of the present application;

[0075] Figure 7This is a fifth flow chart of the iron tapping dust removal control method based on a multimodal fusion perception agent in an embodiment of the present application;

[0076] Figure 8 This is a sixth flow chart of the iron tapping dust removal control method based on a multimodal fusion perception agent in an embodiment of the present application;

[0077] Figure 9 This is one of the structural diagrams of the iron tapping dust removal control device based on the multimodal fusion perception intelligent agent in the embodiment of the present application;

[0078] Figure 10 This is the second structural diagram of the iron tapping dust removal control device based on the multimodal fusion perception intelligent agent in the embodiment of the present application;

[0079] Figure 11 This is the third structural diagram of the iron tapping dust removal control device based on the multimodal fusion perception intelligent agent in the embodiment of the present application;

[0080] Figure 12 This is the fourth structural diagram of the iron tapping dust removal control device based on the multimodal fusion perception intelligent agent in the embodiment of the present application;

[0081] Figure 13 This is the fifth structural diagram of the iron tapping dust removal control device based on the multimodal fusion perception intelligent agent in the embodiment of the present application;

[0082] Figure 14 The sixth structural diagram of the iron-out dust removal control device based on multimodal fusion perception intelligent agent in the embodiment of this application

[0083] Figure 15 This is the seventh structural diagram of the iron tapping dust removal control device based on the multimodal fusion perception intelligent agent in the embodiment of the present application;

[0084] Figure 16 Schematic diagram of the structure of the electronic device in the embodiment of the present application. DETAILED DESCRIPTION

[0085] To make the purpose, technical solutions, and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0086] The acquisition, storage, use, and processing of data in this application's technical solution comply with relevant national laws and regulations.

[0087] In order to use multimodal fusion perception intelligent agent technology to achieve real-time monitoring of working condition changes and smoke dust concentration during blast furnace iron tapping and intelligent control of dust removal equipment, the intelligent agent constructed by deep reinforcement learning algorithm can automatically adjust the control strategy of dust removal equipment, improve dust removal effect and reduce equipment energy consumption. This application provides an embodiment of a method for controlling iron tapping dust removal based on multimodal fusion perception intelligent agent, see Figure 1 The iron tapping dust removal control method based on multimodal fusion perception intelligent agent specifically includes the following contents:

[0088] Step S101: collecting environmental parameters of the iron-making site and operating parameters of the dust removal equipment at fixed time intervals.

[0089] First, environmental parameters at the blast furnace tapping site (such as smoke concentration and temperature) and operating parameters of the dust removal equipment (such as speed and valve opening) are automatically collected at regular intervals. This ensures real-time monitoring and data continuity, providing a foundation for subsequent analysis.

[0090] Step S102: extracting features from the environmental state parameters and the dust removal equipment action parameters, and constructing an evaluation network and a target network based on the feature values.

[0091] Next, data processing techniques are used to extract features from the collected parameters, such as analyzing video data through image recognition or performing numerical analysis on sensor data. Based on these feature values, two neural network models are constructed: an evaluation network and a target network. These models analyze the data and predict the optimal dust removal strategy.

[0092] Step S103: Based on a deep reinforcement learning algorithm, an optimization model of the iron tapping dust removal control strategy is obtained through a reinforcement learning process of the interaction between the evaluation network and the target network training agent and the environment.

[0093] The Deep Reinforcement Learning (DQN) algorithm can be applied to train the intelligent agent through an evaluation network and a target network, enabling it to learn how to interact with the environment in a simulated or real iron-making site. In this process, the intelligent agent learns to identify the most effective dust removal control strategy through trial and error, forming an optimized control model.

[0094] Step S104: applying the optimization model to the dust removal equipment to perform dust removal control on the iron-making site.

[0095] Finally, this optimization model was applied to actual dust removal equipment, enabling it to automatically adjust operating parameters based on the real-time environment and equipment status. This not only improved dust removal efficiency but also helped reduce energy consumption and labor costs, achieving more efficient and intelligent environmental management.

[0096] From the above description, it can be seen that the iron-making dust removal control method based on multimodal fusion perception intelligent body provided in the embodiment of the present application can realize real-time monitoring of operating condition changes and smoke dust concentration during the blast furnace iron-making process and intelligent control of dust removal equipment, automatically adjust the control strategy of dust removal equipment, improve dust removal effect and reduce equipment energy consumption.

[0097] In one embodiment of the iron tapping dust removal control method based on multimodal fusion perception intelligent agent of the present application, see Figure 2 The environmental state parameters collected at the iron-making site include:

[0098] Step S101A: collecting smoke and dust images of the current iron-making environment through a camera;

[0099] Step S101B: Collect smoke data of the current iron-making environment through the total suspended particulate matter sensor and the laser sensor.

[0100] In an optional embodiment, the operating parameters of the dust removal equipment at the iron-making site include the dust removal fan speed and valve opening.

[0101] First, the data acquisition program is continuously run during the production process to collect high-definition camera video frames IMG, TSP sensor data, and VDM sensor data at fixed time intervals.

[0102] During the data collection process, the dust collector speed R and the specified valve opening Valve need to be recorded simultaneously. The control action time defaults to a fixed time interval Δt. The above three variables constitute the action operation space A of the dust removal control strategy. t :

[0103] A t =(VALVE t ,R t ,Δt)

[0104] All of the above data will be aligned according to the time dimension, and the constructed environment and action simulation playback pool ExperiencePool will be described as follows:

[0105] ExperiencePool = {(S t ,A t ,S t+Δt ,Reward t )}

[0106] The above is a description of the structure of the agent training data.

[0107] Preferably, during the actual data collection process, the environmental status data of the dust removal site and the sampling actions in the operation record data are repeatedly recorded at a fixed time interval Δt until sufficient environmental data is collected. The environmental interaction data is then recorded and integrated according to preset rules. Environmental feedback data should be collected at multiple speeds and valve opening values ​​in the field, covering as many scenario combinations as possible and collecting as much data as possible to diversify the training data. If scenario coverage is incomplete, the time required for model training to converge will be greatly increased.

[0108] Environmental status data is mainly based on sensors, and it is sufficient to ensure coverage of all scenarios in the training set.

[0109] The data for the control action include the dust removal fan speed and valve opening. Both are continuous variables. Here we discretize them based on experience. The basic principle is to perform empirical uniform sampling within the range of each action value. The details are as follows:

[0110] R t ∈{1000,3000,5000,7000,9000,11000,13000,15000}, unit is r / min;

[0111] VALVE t ∈{0.1,0.3,0.5,0.7,0.9,1}

[0112] The above values ​​are combined in pairs, and a fixed action time Δt is added to form the action space A. The enumeration of all action sets is as follows:

[0113] A={(1000,0.1,Δt),(1000,0.4,Δt),……(2000,0.1,Δt)……}

[0114] Based on the above steps, the combined use of cameras and various sensors (such as total suspended particulate matter sensors and laser sensors) enables more comprehensive and accurate monitoring of the dust and smoke conditions in the iron-making environment. Based on real-time collected environmental parameters and dust removal equipment operating parameters (such as fan speed and valve opening), dust removal strategies can be adjusted more flexibly and precisely, thereby improving dust removal efficiency and adaptability.

[0115] In one embodiment of the iron tapping dust removal control method based on multimodal fusion perception intelligent agent of the present application, see Figure 3 The feature extraction of the environmental state parameters and the dust removal equipment action parameters may further specifically include the following contents:

[0116] Step S102A: extracting features from the smoke image based on an image encoder;

[0117] Step S102B: constructing a complete text description based on the smoke data, extracting features from the text description using a text encoder, and dimensionally fusing features of the smoke image with features of the smoke data;

[0118] Step S102C: extracting features of the dust removal equipment action parameters based on the text encoder.

[0119] like Figure 4 As shown, in this application, the Clip model is used to extract features from image and text data. The image data is subjected to feature extraction through the image editor Image encoder, which means that the system can identify and abstract useful information from the image. For sensor data such as TSP (total suspended particulate matter) and VDM (video monitoring data), as well as data on operating actions (such as fan speed and valve opening and closing), these non-image data need to be converted into text descriptions first. Then, these text descriptions are input into the text editor text encoder for feature extraction to capture the key information in the sensor data and operating actions. Finally, the features of the image and text are fused in dimension to build a model that fully understands the current iron-making environment. The above fusion method not only enhances the model's ability to understand multiple data types, but also improves the accuracy and efficiency of decision-making.

[0120] The fusion method of the above three types of data adopts a feature extraction method based on the Clip pre-trained model; among them, the image data uses the corresponding Image Encoder solution for feature extraction; the VDM and TSP sensor data and action strategy information first use a fixed template to construct a complete text description (prompt), and then use the Text Encoder method for feature processing. This solution effectively solves the problems of sensor information being too low in dimension compared to image data, unbalanced information expression, and consistency in the fusion of data in different formats.

[0121] In one embodiment of the iron tapping dust removal control method based on multimodal fusion perception intelligent agent of the present application, see Figure 5 The reinforcement learning process of training the agent to interact with the environment through the evaluation network and the target network based on the deep reinforcement learning algorithm may further specifically include the following contents:

[0122] Step S103A: Initializing the evaluation network and the target network so that the evaluation network and the target network have consistent structures and independent parameters;

[0123] Step S103B: In each training cycle, use the experience replay mechanism to randomly sample from historical interactions to reduce the correlation and variance in training;

[0124] Step S103C: Based on the temporal difference update rule, the weights of the evaluation network are adjusted after each training cycle according to the reward signal and the maximum expected reward of the next state to select the execution action that can maximize the expected reward.

[0125] Among them, the Q function constructed in this application adopts a neural network model, with the environment state and action S t ,A t as the input of the network.

[0126] Evaluation Network The target network Q adopts the same network structure as shown in the figure.

[0127] Construct the optimization objective function. The optimization function continues to use the temporal difference evaluation method, namely the Bellman equation, as follows:

[0128]

[0129] Among them, the hyperparameter γ represents the penalty coefficient in the interval [0,1]. The larger the value, the more attention is paid to future returns; conversely, the more attention is paid to current returns.

[0130] If the on-site environment meets the standards, the dust removal will be terminated (i.e. t =1), then y t =reward t .

[0131] Among them, the calculation result y t To evaluate the network The target value to be fitted, the mean square error (MSE) is used as the loss function, so that The network continuously updates its weights in subsequent training rounds to achieve evolution; and the target network Q periodically accumulates a certain number of iterations and then replaces its own parameters with the latest ones. The parameters of the two networks are kept consistent and iterated synchronously in this way.

[0132] For the agent learning module, the Q network is modified based on the original DQN neural network, adding an environmental feature extraction link to fuse three types of sensor data

[0133] like Figure 6 As shown, the above steps are repeated for multiple rounds of training iterations. In each round, the agent interacts with the environment, collects new experience samples, and uses these samples to update the parameters of the Q network. At the same time, the parameters of the target network are regularly updated to gradually approach the parameters of the current Q network.

[0134] During the actual training process, reinforcement learning relies on the feedback of the actual environment to select the action strategy for the next stage. However, due to the complexity of the industrial environment, an experimental test environment needs to be configured on site to collect environmental feedback information in a timely manner. t On the other hand, the control strategy of the intelligent agent can be evaluated by observing the dust removal effect on site. The overall training process can be separated into two parts: offline and online.

[0135] First, the Q network is updated for the first time using offline training using the full offline data. This process is based entirely on historical observations and requires only offline execution. Field verification has shown that this approach effectively improves the convergence rate of the Q network parameters.

[0136] Next, based on the latest Q network, various system modules are deployed on site and the PLC control system is introduced for real-world training and iteration. This training process uses the experience replay technology to store and reuse the experience data generated during the real-world training process. At each time step, the agent will store the current environment state S t , Action A t 、Reward t and the new state s t+1 The experience samples are randomly drawn from this buffer during training.

[0137] In one embodiment of the iron tapping dust removal control method based on a multimodal fusion perception agent of the present application, the following contents may also be specifically included:

[0138] A reward function is constructed with the maximum dust removal effect and the minimum energy consumption of the dust removal equipment as reward indicators, where:

[0139] In response to a decrease in smoke concentration detected by the sensor, the reward increases;

[0140] In response to energy consumption resulting from operation of the device exceeding a predetermined threshold, the reward is reduced.

[0141] First, the data acquisition program is continuously run during the production process, collecting high-definition camera video frames IMG, TSP sensor data, VDM sensor data, and fan operating power P at fixed time intervals. Based on the above observation indicators, the Terminate variable is calculated to determine whether the intelligent agent control behavior begins to intervene. The above indicators together constitute the environmental state set S at time t t :

[0142] S t =(IMG t ,TSP t,VDM t ,Terminate t ,P t )

[0143] The value range of the Terminate variable is {0, 1}. The value is determined by the OR operation of the iron tapping signal Out, TSP, and VDM, that is:

[0144]

[0145] The power does not directly affect the dust removal environment, but is only used as an energy consumption analysis factor, so it is only used to calculate the state reward value. t .

[0146] The state space and action space of the agent have been constructed, and the construction and calculation of the reward function depends on the current state space value and the motor power energy consumption index. In general, when constructing a reward function based on dust removal effect and equipment energy consumption, two main goals need to be considered: one is to maximize the dust removal effect, and the other is to minimize the equipment energy consumption. There is a certain conflict between these two goals, so it is necessary to use a weighted average coefficient in the reward function to make appropriate trade-offs. The formula for defining the reward function in this scenario in this application is as follows:

[0147]

[0148] The above formula satisfies the condition w1+w2=1, w 11 +w 12 =1

[0149] All weights in the above formula are greater than 0. Among them, the dust removal effect and energy consumption are weighted averaged by weight coefficients w1 and w2 respectively, w 11 and w 12 It is used to weigh the test results of TSP sensor and VDM sensor respectively. Here, the product of power and time interval is used to approximate the energy consumption of the device between two sampling intervals, that is, p t Δt.

[0150] For the weight coefficient, if the agent needs to focus more on energy saving in its optimization direction, then the energy penalty coefficient w2 should be appropriately increased; if the TSP observation results are more important, then w 11 value.

[0151] In one embodiment of the iron tapping dust removal control method based on multimodal fusion perception intelligent agent of the present application, see Figure 7 , and can also include the following:

[0152] Step S103D: In response to the agent decision process starting, setting an exploration rate value, and selecting a random action from the entire action set at the exploration rate;

[0153] Step S103E: During the training process, the exploration rate is lowered to reduce random exploration behavior, and a strategy for reducing the exploration rate is determined based on the learning progress of the agent and environmental feedback.

[0154] During training, the target network Q updates its actions at a fixed time interval Δt and uses an ε-greedy strategy to select an action strategy. This strategy selects a random action from the full set of actions A with probability ε and selects the action that maximizes the target network Q value with probability 1-ε. As training progresses, the value of the control ε is gradually reduced, causing the agent to select more actions with the largest Q value.

[0155] The overall training process needs to be carried out in batches multiple times in the experimental environment. At the same time, in the offline environment, high-quality strategies in the strategy pool can be used for offline training and learning to accelerate the convergence efficiency of the intelligent agent.

[0156] This application makes full use of the improved DQN model architecture and training strategy, including strategy pool cache management, dual-Q neural network optimization, ε-greedy strategy, etc., for the construction and training process of the dust removal agent.

[0157] In one embodiment of the iron tapping dust removal control method based on multimodal fusion perception intelligent agent of the present application, see Figure 8 , and can also include the following:

[0158] Step S104A: dynamically adjusting the action parameters of the dust removal equipment according to the control strategy output by the optimization model;

[0159] Step S104B: Regularly evaluate the overall performance of the dust removal system. If it is detected that the dust removal effect and equipment energy consumption do not meet the expected indicators, feedback is given to the next reinforcement learning process.

[0160] After training is complete, the model parameters are fixed, the final model is output, and the model is deployed in a production environment. The performance of the agent is evaluated by running it in a real-world test environment. In this application, the learning effect is measured based on the total reward ∑Reward.

[0161] Furthermore, in this application, the maturity and stability of the intelligent agent are achieved through multiple iterations of training and learning. After each round of offline training, the updated intelligent agent can be deployed in the production environment to collect new environmental data through interaction with the real production process. This new data is then used to construct a new training set for the next round of training, thereby continuously improving the performance of the intelligent agent. This process can be repeated until the intelligent agent can well adapt to and optimize the dust removal effect in the production environment.

[0162] In order to utilize multimodal fusion perception intelligent agent technology to realize real-time monitoring of working condition changes and smoke dust concentration during blast furnace iron tapping and intelligent control of dust removal equipment, the intelligent agent constructed by deep reinforcement learning algorithm can automatically adjust the control strategy of dust removal equipment, improve dust removal effect and reduce equipment energy consumption. This application provides an embodiment of a multimodal fusion perception intelligent agent-based iron tapping dust removal control device for realizing all or part of the contents of the multimodal fusion perception intelligent agent-based iron tapping dust removal control method, see Figure 9 The iron tapping dust removal control device based on multimodal fusion perception intelligent agent specifically includes the following contents:

[0163] The multimodal environment perception module 1101 is used to collect environmental state parameters of the iron-making site and dust removal equipment operation parameters at fixed time intervals;

[0164] The feature extraction module 1102 is used to extract features of the environmental state parameters and the dust removal equipment action parameters, and construct an evaluation network and a target network based on the feature values;

[0165] The agent learning module 1103 is configured to: based on a deep reinforcement learning algorithm, train the agent through the evaluation network and the target network to conduct a reinforcement learning process of interaction with the environment, thereby obtaining an optimization model for the iron tapping dust removal control strategy;

[0166] The intelligent agent decision module 1104 is used to apply the optimization model to the dust removal equipment to perform dust removal control on the iron-making site.

[0167] According to any embodiment of this application, see Figure 10 , the agent learning module includes:

[0168] The image acquisition unit 1101A is used to: acquire smoke and dust images of the current iron-making environment through a camera;

[0169] The data acquisition unit 1101B is used to collect smoke data of the current iron-making environment through a total suspended particulate matter sensor and a laser sensor.

[0170] According to any embodiment of the present application, the operating parameters of the dust removal equipment at the iron-making site include the dust removal fan speed and the valve opening.

[0171] According to any embodiment of this application, see Figure 11 , the feature extraction module includes:

[0172] The image feature unit 1102A is configured to extract features from the smoke image based on an image encoder;

[0173] The data feature unit 1102B is configured to construct a complete text description based on the smoke data, extract features from the text description using a text encoder, and perform dimension fusion on features of the smoke image and features of the smoke data.

[0174] The action feature unit 1102C is used to extract features of the action parameters of the dust removal equipment based on the text encoder.

[0175] According to any embodiment of this application, see Figure 12 , the agent learning module includes:

[0176] Initialization unit 1103A, configured to initialize the evaluation network and the target network so that the evaluation network and the target network have consistent structures and independent parameters;

[0177] The training unit 1103B is configured to: randomly sample from historical interactions using an experience replay mechanism in each training cycle to reduce correlation and variance in training;

[0178] The optimization unit 1103C is used to: adjust the weights of the evaluation network after each training cycle based on the temporal difference update rule, according to the reward signal and the maximum expected benefit of the next state, so as to select an execution action that can maximize the expected reward.

[0179] According to any embodiment of the present application, it further includes a reward building module for:

[0180] A reward function is constructed with the maximum dust removal effect and the minimum energy consumption of the dust removal equipment as reward indicators, where:

[0181] In response to a decrease in smoke concentration detected by the sensor, the reward increases;

[0182] In response to energy consumption resulting from operation of the device exceeding a predetermined threshold, the reward is reduced.

[0183] According to any embodiment of this application, see Figure 13 , also includes greedy optimization modules, including:

[0184] An exploration selection unit 1103D is configured to: in response to the agent decision process starting, set an exploration rate value, and select a random action from the entire action set at the exploration rate;

[0185] The exploration optimization unit 1103E is used to: reduce the exploration rate during the training process to reduce random exploration behavior, and determine the exploration rate reduction strategy based on the learning progress of the intelligent agent and environmental feedback.

[0186] According to any embodiment of this application, see Figure 14 , the agent decision module includes:

[0187] The control output unit 1104A is used to dynamically adjust the action parameters of the dust removal equipment according to the control strategy output by the optimization model;

[0188] The feedback improvement unit 1104B is used to: regularly evaluate the overall performance of the dust removal system, and if it is detected that the dust removal effect and equipment energy consumption do not meet the expected indicators, feedback is given to the next reinforcement learning process.

[0189] From the above description, it can be seen that the iron-making dust removal control device based on multimodal fusion perception intelligent body provided in the embodiment of the present application can realize real-time monitoring of operating condition changes and smoke dust concentration during the blast furnace iron-making process and intelligent control of the dust removal equipment, automatically adjust the control strategy of the dust removal equipment, improve the dust removal effect and reduce equipment energy consumption.

[0190] To further illustrate this solution, the present application also provides a specific application example of using the above-mentioned iron-out dust removal control device based on multimodal fusion perception intelligent agent to implement the iron-out dust removal control device based on multimodal fusion perception intelligent agent, which specifically includes the following contents:

[0191] Multimodal Environmental Perception Module: This module is also known as the environmental data acquisition module. It uses sensors to monitor total suspended particulate matter (TSP) concentration near the blast furnace tapping process, smoke sensor parameters such as VDM laser sensors, and other parameters to obtain environmental information. Furthermore, fixed cameras monitor smoke emissions from the blast furnace tapping hole in real time, acquiring visual data on smoke emissions.

[0192] Agent Learning Module: This module is responsible for designing and training the decision-making model structure. Based on a modified Deep Reinforcement Learning (DQN) algorithm, this module builds an agent capable of sensing changes in operating conditions and automatically adjusting control strategies. By interactively learning with the offline environment, the agent continuously optimizes the control strategy to improve dust removal effectiveness and reduce equipment energy consumption.

[0193] Control strategy optimization module: Based on the optimization control model output by the intelligent agent learning module, the operating parameters of the dust removal equipment, such as fan speed and the opening of the dust collector inlet and outlet valves, are adjusted at fixed time intervals to achieve intelligent control of the dust removal equipment.

[0194] like Figure 15 As shown, during the blast furnace iron-making process, the multimodal perception module monitors environmental changes and smoke emissions in real time, and passes the relevant information to the intelligent learning module. Based on the received information, the intelligent learning module performs training and learning in an offline environment through feature extraction and deep reinforcement learning algorithm (DQN), and outputs the optimized control strategy. The control strategy optimization module dynamically adjusts the operating parameters of the dust removal equipment according to the output results of the intelligent learning module to realize intelligent control of the dust removal process. By continuously optimizing the control strategy, this application can improve the dust removal effect and reduce the energy consumption of the equipment.

[0195] From a hardware perspective, in order to utilize multimodal fusion perception agent technology to achieve real-time monitoring of operating condition changes and smoke dust concentration during blast furnace tapping and intelligent control of dust removal equipment, an agent constructed using a deep reinforcement learning algorithm can automatically adjust the control strategy of the dust removal equipment, improve the dust removal effect, and reduce equipment energy consumption. This application provides an embodiment of an electronic device for implementing all or part of the contents of the iron tapping dust removal control method based on a multimodal fusion perception agent. The electronic device specifically includes the following contents:

[0196] Processor (processor), memory (memory), communication interface (Communications Interface) and bus; wherein, the processor, memory, and communication interface communicate with each other through the bus; the communication interface is used to realize information transmission between the iron-outlet dust removal control device based on the multimodal fusion perception intelligent body and related equipment such as the core business system, user terminal and related database; the logic controller can be a desktop computer, a tablet computer and a mobile terminal, etc., and this embodiment is not limited to this. In this embodiment, the logic controller can be implemented with reference to the embodiment of the iron-outlet dust removal control method based on the multimodal fusion perception intelligent body and the embodiment of the iron-outlet dust removal control device based on the multimodal fusion perception intelligent body in the embodiment, and the contents thereof are merged here, and the repeated parts are not repeated.

[0197] It is understandable that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0198] In practical applications, part of the iron-out dust removal control method based on the multimodal fusion perception agent can be executed on the electronic device side as described above, or all operations can be completed in the client device. The specific selection can be based on the processing power of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed in the client device, the client device may also include a processor.

[0199] The client device may include a communication module (i.e., a communication unit) that can establish a communication connection with a remote server to implement data transmission with the server. The server may include a server on the task scheduling center side, and in other implementation scenarios, may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0200] Figure 16 Schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Figure 16 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that the Figure 16 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0201] In one embodiment, the function of the iron tapping dust removal control method based on the multimodal fusion perception agent can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0202] Step S101: collecting environmental parameters of the iron-making site and operating parameters of the dust removal equipment at fixed time intervals;

[0203] Step S102: extracting features of the environmental state parameters and the dust removal equipment action parameters, and constructing an evaluation network and a target network based on the feature values;

[0204] Step S103: Based on a deep reinforcement learning algorithm, an optimization model of the iron tapping dust removal control strategy is obtained through a reinforcement learning process of the interaction between the evaluation network and the target network training agent and the environment;

[0205] Step S104: applying the optimization model to the dust removal equipment to remove dust from the iron tapping site.

[0206] From the above description, it can be seen that the electronic equipment provided in the embodiment of the present application realizes real-time monitoring of operating condition changes and smoke concentration during blast furnace iron tapping and intelligent control of dust removal equipment, automatically adjusts the control strategy of dust removal equipment, improves dust removal effect and reduces equipment energy consumption.

[0207] In another embodiment, the iron-out dust removal control device based on the multimodal fusion perception intelligent agent can be configured separately from the central processing unit 9100. For example, the iron-out dust removal control device based on the multimodal fusion perception intelligent agent can be configured as a chip connected to the central processing unit 9100, and the function of the iron-out dust removal control method based on the multimodal fusion perception intelligent agent can be realized through the control of the central processing unit.

[0208] like Figure 16 As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Figure 16 In addition, the electronic device 9600 may also include all components shown in Figure 16 For components not shown, reference may be made to the prior art.

[0209] like Figure 16 As shown, the central processing unit 9100 is sometimes also referred to as a controller or operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0210] Memory 9140 can be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It can store the aforementioned failure-related information and also store programs that execute the relevant information. The CPU 9100 can execute the programs stored in memory 9140 to implement information storage or processing.

[0211] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 may be, for example, a keypad or touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display objects such as images and text. The display may be, for example, an LCD display, but is not limited thereto.

[0212] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), or a SIM card. Alternatively, it may be a memory that retains information even when power is off, can be selectively erased, and is provided with more data. Examples of such memory are sometimes referred to as EPROMs. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 by the central processing unit 9100.

[0213] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various driver programs for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0214] The communication module 9110 is a transmitter / receiver 9110 that transmits and receives signals via an antenna 9111. The communication module (transmitter / receiver) 9110 is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as in a conventional mobile communication terminal.

[0215] Based on different communication technologies, multiple communication modules 9110 can be provided in the same electronic device, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module. The communication module (transmitter / receiver) 9110 is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby implementing common telecommunication functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Furthermore, the audio processor 9130 is also coupled to the central processing unit 9100, enabling local recording via the microphone 9132 and playback of stored audio via the speaker 9131.

[0216] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all steps of the method for controlling dust removal during iron tapping based on a multimodal fusion perception agent, in which the execution subject is a server or a client in the above-mentioned embodiment. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, all steps of the method for controlling dust removal during iron tapping based on a multimodal fusion perception agent, in which the execution subject is a server or a client in the above-mentioned embodiment, are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0217] Step S101: collecting environmental parameters of the iron-making site and operating parameters of the dust removal equipment at fixed time intervals;

[0218] Step S102: extracting features of the environmental state parameters and the dust removal equipment action parameters, and constructing an evaluation network and a target network based on the feature values;

[0219] Step S103: Based on a deep reinforcement learning algorithm, an optimization model of the iron tapping dust removal control strategy is obtained through a reinforcement learning process of the interaction between the evaluation network and the target network training agent and the environment;

[0220] Step S104: applying the optimization model to the dust removal equipment to remove dust from the iron tapping site.

[0221] From the above description, it can be seen that the computer-readable storage medium provided in the embodiment of the present application realizes real-time monitoring of operating condition changes and smoke concentration during blast furnace iron tapping and intelligent control of dust removal equipment, automatically adjusts the control strategy of dust removal equipment, improves dust removal effect and reduces equipment energy consumption.

[0222] The embodiments of the present application also provide a computer program product capable of implementing all steps of the iron tapping dust removal control method based on a multimodal fusion perception intelligent agent in the above-mentioned embodiment, where the execution subject is a server or a client. When the computer program / instruction is executed by a processor, the steps of the iron tapping dust removal control method based on a multimodal fusion perception intelligent agent are implemented. For example, the computer program / instruction implements the following steps:

[0223] Step S101: collecting environmental parameters of the iron-making site and operating parameters of the dust removal equipment at fixed time intervals;

[0224] Step S102: extracting features of the environmental state parameters and the dust removal equipment action parameters, and constructing an evaluation network and a target network based on the feature values;

[0225] Step S103: Based on a deep reinforcement learning algorithm, an optimization model of the iron tapping dust removal control strategy is obtained through a reinforcement learning process of the interaction between the evaluation network and the target network training agent and the environment;

[0226] Step S104: applying the optimization model to the dust removal equipment to remove dust from the iron tapping site.

[0227] From the above description, it can be seen that the computer program product provided in the embodiment of the present application realizes real-time monitoring of operating condition changes and smoke concentration during blast furnace iron-making and intelligent control of dust removal equipment, automatically adjusts the control strategy of dust removal equipment, improves dust removal effect and reduces equipment energy consumption.

[0228] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0229] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0230] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0231] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0232] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. A method for controlling iron tapping dust removal based on a multimodal fusion perception agent, characterized in that: The method comprises: Collect environmental parameters of the iron-making site and dust removal equipment operating parameters at fixed time intervals; Extracting features of the environmental state parameters and the dust removal equipment action parameters, and constructing an evaluation network and a target network based on the feature values; Based on the deep reinforcement learning algorithm, an optimization model of the iron tapping dust removal control strategy is obtained through the reinforcement learning process of the evaluation network and the target network training agent interacting with the environment; The optimization model is applied to the dust removal equipment to perform dust removal control on the iron-making site.

2. The iron tapping dust removal control method based on multimodal fusion perception intelligent agent according to claim 1 is characterized in that: The environmental state parameters collected at the iron-making site include: Use the camera to collect smoke and dust images of the current iron-making environment; The smoke data of the current iron-making environment is collected through the total suspended particulate matter sensor and laser sensor.

3. The iron tapping dust removal control method based on multimodal fusion perception agent according to claim 2 is characterized in that: The operating parameters of the dust removal equipment at the iron-making site include the dust removal fan speed and valve opening.

4. The iron tapping dust removal control method based on multimodal fusion perception agent according to claim 3 is characterized in that: The feature extraction of the environmental state parameters and the dust removal equipment action parameters includes: performing feature extraction on the smoke image based on an image encoder; Constructing a complete text description based on the smoke data, extracting features from the text description using a text encoder, and dimensionally fusing features of the smoke image with features of the smoke data; Feature extraction is performed on the action parameters of the dust removal equipment based on the text encoder.

5. The iron tapping dust removal control method based on multimodal fusion perception intelligent agent according to claim 1 is characterized in that: The reinforcement learning process of training the agent to interact with the environment through the evaluation network and the target network based on the deep reinforcement learning algorithm includes: Initializing the evaluation network and the target network so that the evaluation network and the target network have consistent structures and independent parameters; In each training cycle, an experience replay mechanism is used to randomly sample from historical interactions to reduce correlation and variance in training; Based on the temporal difference update rule, the weights of the evaluation network are adjusted after each training cycle according to the reward signal and the maximum expected benefit of the next state to select the execution action that maximizes the expected reward.

6. The iron tapping dust removal control method based on multimodal fusion perception intelligent agent according to claim 5 is characterized in that: Also includes: A reward function is constructed with the maximum dust removal effect and the minimum energy consumption of the dust removal equipment as reward indicators, where: In response to a decrease in smoke concentration detected by the sensor, the reward increases; In response to energy consumption resulting from operation of the device exceeding a predetermined threshold, the reward is reduced.

7. The iron tapping dust removal control method based on multimodal fusion perception intelligent agent according to claim 5 is characterized in that: Also includes: In response to the agent decision process starting, setting an exploration rate value, and selecting a random action from the entire set of actions at the exploration rate; The exploration rate is lowered during training to reduce random exploration behavior, and the exploration rate reduction strategy is determined based on the learning progress of the agent and environmental feedback.

8. The iron tapping dust removal control method based on multimodal fusion perception intelligent agent according to claim 1 is characterized in that: The applying the optimization model to the dust removal equipment to perform dust removal control on the iron tapping site includes: Dynamically adjusting the action parameters of the dust removal equipment according to the control strategy output by the optimization model; The overall performance of the dust removal system is evaluated regularly. If it is detected that the dust removal effect and equipment energy consumption do not meet the expected indicators, feedback will be provided to the next reinforcement learning process.

9. A dust removal control device for iron tapping based on multimodal fusion perception intelligent agent, characterized in that: The device comprises: The multimodal environmental perception module is used to collect environmental parameters of the ironworks site and the operating parameters of the dust removal equipment at fixed time intervals. A feature extraction module is used to extract features of the environmental state parameters and the dust removal equipment action parameters, and to construct an evaluation network and a target network based on the feature values; An agent learning module is used to: based on a deep reinforcement learning algorithm, train the reinforcement learning process of the agent's interaction with the environment through the evaluation network and the target network to obtain an optimization model of the iron tapping dust removal control strategy; The intelligent agent decision module is used to apply the optimization model to the dust removal equipment to perform dust removal control on the iron-making site.

10. The iron tapping dust removal control device based on multimodal fusion perception intelligent agent according to claim 9, characterized in that: The agent learning module includes: The image acquisition unit is used to: acquire smoke and dust images of the current iron-making environment through a camera; The data acquisition unit is used to collect smoke data of the current iron-making environment through a total suspended particulate matter sensor and a laser sensor.

11. The iron tapping dust removal control device based on multimodal fusion perception intelligent agent according to claim 10, characterized in that: The operating parameters of the dust removal equipment at the iron-making site include the dust removal fan speed and valve opening.

12. The iron tapping dust removal control device based on multimodal fusion perception intelligent agent according to claim 11, characterized in that: The feature extraction module includes: An image feature unit, configured to extract features from the smoke image based on an image encoder; a data feature unit, configured to construct a complete text description based on the smoke data, extract features of the text description using a text encoder, and perform dimension fusion on features of the smoke image and features of the smoke data; An action feature unit is used to extract features of the action parameters of the dust removal equipment based on the text encoder.

13. The iron tapping dust removal control device based on multimodal fusion perception agent according to claim 9, characterized in that: The agent learning module includes: an initialization unit, configured to initialize the evaluation network and the target network so that the evaluation network and the target network have consistent structures and independent parameters; The training unit is used to: randomly sample from historical interactions using the experience replay mechanism in each training cycle to reduce the correlation and variance in training; The optimization unit is used to adjust the weights of the evaluation network after each training cycle based on the reward signal and the maximum expected benefit of the next state based on the temporal difference update rule to select the execution action that can maximize the expected reward.

14. The iron tapping dust removal control device based on multimodal fusion perception intelligent agent according to claim 13, characterized in that: Also included are reward building blocks for: A reward function is constructed with the maximum dust removal effect and the minimum energy consumption of the dust removal equipment as reward indicators, where: In response to a decrease in smoke concentration detected by the sensor, the reward increases; In response to energy consumption resulting from operation of the device exceeding a predetermined threshold, the reward is reduced.

15. The iron tapping dust removal control device based on multimodal fusion perception intelligent agent according to claim 13, characterized in that: Also includes greedy optimization modules, including: an exploration selection unit, configured to: in response to the agent decision process starting, set an exploration rate value, and select a random action from the entire action set at the exploration rate; The exploration optimization unit is used to reduce the exploration rate during training to reduce random exploration behavior, and determine the exploration rate reduction strategy based on the learning progress of the agent and environmental feedback.

16. The iron tapping dust removal control device based on multimodal fusion perception intelligent agent according to claim 9, characterized in that: The agent decision module includes: A control output unit, configured to dynamically adjust the action parameters of the dust removal equipment according to the control strategy output by the optimization model; The feedback improvement unit is used to regularly evaluate the overall performance of the dust removal system. If it is detected that the dust removal effect and equipment energy consumption do not meet the expected indicators, feedback will be provided to the next reinforcement learning process.

17. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the iron tapping dust removal control method based on a multimodal fusion perception agent according to any one of claims 1 to 8 are implemented.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the iron tapping dust removal control method based on a multimodal fusion perception intelligent agent as described in any one of claims 1 to 8 are implemented.

19. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the steps of the iron tapping dust removal control method based on a multimodal fusion perception intelligent agent as described in any one of claims 1 to 8 are implemented.

Images